Google Confirms Using YouTube Videos to Train Veo 3 AI

Google is using a portion of YouTube’s vast video library to train Veo 3, its latest AI-powered video generation tool, the company confirmed to CNBC.

While only a subset of content is used, the practice has raised concerns among creators, many of whom say they were unaware their videos could be used to train AI models that may eventually compete with them.

YouTube allows creators to block third-party companies from using their content for AI training—but not Google itself.

Other tech giants like Apple, Anthropic, Amazon, Meta, and Microsoft also access YouTube data for AI model training.

In a blog post published in September, the company did state that YouTube content may be used to “improve the product experience … including through machine learning and AI applications.”

Users who upload videos to the platform have no option to opt out of allowing Google to use their content for AI training.

Veo 3, unveiled at Google I/O 2025, is the company’s third-generation video generation model.

It can create eight-second videos and now includes audio generation—a feature not yet available in rivals like OpenAI’s Sora.

Initially released in the U.S., Veo 3 is now accessible in over 70 countries. Users can access the tool via Google’s $20/month Gemini Pro or $249/month Gemini Ultra plans.